Decision Making during the Psychological Refractory Period Ariel Zylberberg, Brian Ouellette, Mariano Sigman, Pieter R. Roelfsema  Current Biology  Volume 22, Issue 19, Pages 1795-1799 (October 2012) DOI: 10.1016/j.cub.2012.07.043 Copyright © 2012 Elsevier Ltd Terms and Conditions

Current Biology 2012 22, 1795-1799DOI: (10.1016/j.cub.2012.07.043) Copyright © 2012 Elsevier Ltd Terms and Conditions

Figure 1 Models of Evidence Accumulation during the PRP and Design of the Experiment to Distinguish between Them (A) A decision is made by the accumulation of sensory evidence. Noisy sensory evidence is shown in gray, and the accumulated evidence (Acc) for T1 and T2 is shown in light blue and red, respectively. For clarity, only a single barrier is shown, but two symmetrical bounds at ±B0 are generally used to model two-alternative forced-choice decisions. The accumulation process ends when it reaches one of these barriers. Additional perceptual (P) and postaccumulation (PAcc) latencies add to the total response time. Three alternative models are shown for the second task of the PRP paradigm, which differ in how the accumulation of evidence for T2 is influenced by T1. If only one accumulation process can proceed at a time, then the integration for T2 is delayed at a short SOA (“serial integration”) [8]. Alternatively, if the accumulation of evidence for the two tasks can proceed in parallel, no influence should be observed on the accumulation process (“parallel integration”). Dual-task interference could also be caused by a reduction in the efficacy of accumulation for T2 (“partial integration”). Note that all models can result in the same response-time to T2 (RT2) by adjusting PAcc. (B) Experimental design. Subjects performed a tone discrimination task (T1) followed by a luminance discrimination task (T2), which involved deciding which of two patches was brightest. Each patch consisted of four bars and independent luminance noise was added to each bar and updated at a frequency of 25 Hz. The SOAs were randomly selected on each trial and could take values of [0, 120, 200, 520, 600] ms. Responses were made as fast as possible, using different hands for the two tasks. (C) Magnification of one frame of the visual stimulus. The target patch with highest luminance is on the left. Current Biology 2012 22, 1795-1799DOI: (10.1016/j.cub.2012.07.043) Copyright © 2012 Elsevier Ltd Terms and Conditions

Figure 2 Response Times and Psychophysical Kernels (A) Average response times for the tone (T1) and visual (T2) tasks. Different SOAs are indicated with different colors (see legend). Error bars indicate SEM. (B) Classification images. Each panel corresponds to a different SOA. Two classification images are shown in each panel: the solid line shows the difference in the luminance fluctuations between correct and incorrect trials of the “target” patch. On correct trials, the luminance of the target patch tended to be higher. The dashed line shows the same subtraction for the “distractor” patch, which tended to have a lower luminance on correct trials. (C) Time course of the classification images (“target”–“distractor”), after aligning the luminance fluctuations to the onset of the visual stimulus (left) or to the time of the response (right). Noise averages in the left portion of the graph are drawn until one of the three participants had responded in more than 70% of trials at that SOA. Samples after the subject’s response were excluded from the trial averages. On the right, the fluctuations were aligned to the time of the response before computing the classification images. Noise averages in this portion of the graph are drawn from the times at which every participant has started integration in at least 30% of trials. The classification images of individual participants are shown in Figure S1A. We obtained similar psychophysical kernels with a linear regression analysis (Figure S1B). Current Biology 2012 22, 1795-1799DOI: (10.1016/j.cub.2012.07.043) Copyright © 2012 Elsevier Ltd Terms and Conditions

Figure 3 Fit of a Diffusion Model of Dual-Task Interference (A) Response time distributions as a function of SOA, for the participants (colored bars) and the model (black lines). Response times are shown relative to the onset of the visual stimulus of task 2. The fraction of variance explained by the model fit (R2) is, from left to right, 0.99, 0.98, 0.99, 0.98, and 0.98. Bin width is 40 ms. (B) Classification images for the data (colored lines) and model (black lines), as a function of SOA, normalized to the peak value across SOA. The fraction of variance explained by the model fit (R2) is, from left to right, 0.78, 0.92, 0.91, 0.93, and 0.9. In (A and B), the model was fitted independently for each participant, and the resulting fits were averaged across participants. (C) The model separated RT2 (solid lines) into a time of accumulation (dashed lines, AccT; time at which the boundary is reached) and a postaccumulation time (dotted lines, PAccT). Error bars indicate SEM across participants. Response time distributions and classification images for individual participants are shown in Figure S2, and model parameters are shown in Table S1. Current Biology 2012 22, 1795-1799DOI: (10.1016/j.cub.2012.07.043) Copyright © 2012 Elsevier Ltd Terms and Conditions

Figure 4 Serial Integration of Evidence for Multiple Decisions (A) Subjects had to trace a curve by choosing the brightest luminance directions at three locations. As in our main experiment, the luminance at each bifurcation varied rapidly over time (25 Hz) so that we could measure evidence accumulation for the three decisions. (B) The classification images, aligned to the time of the third response. Evidence for the three decisions accumulated serially, implying that low levels of the visual system do not automatically integrate luminance information for this task. The bottom panel shows the distribution of response times for the three decisions, aligned to the last response. Current Biology 2012 22, 1795-1799DOI: (10.1016/j.cub.2012.07.043) Copyright © 2012 Elsevier Ltd Terms and Conditions